ArticlePDF Available

Unified Framework for Ship Detection in Multi-Frequency SAR Images: A Demonstration with COSMO-SkyMed, Sentinel-1, and SAOCOM Data

MDPI
Remote Sensing
Authors:

Abstract and Figures

In the framework of maritime surveillance, vessel detection techniques based on spaceborne synthetic aperture radar (SAR) images have promoted extensive applications for the effective understanding of unlawful activities at sea. This paper deals with this topic, presenting a novel approach that exploits a cascade application of a pre-screening algorithm and a discrimination phase. Pre-screening is based on a constant false alarm rate (CFAR) detector, whereas discrimination exploits sub-look analysis (SLA). For the first time, the method has been validated with experiments on multi-frequency (C-, X-, and L-band) SAR images, demonstrating a significant reduction of up to 40% in false alarms within highly congested scenarios, along with a notable enhancement of the receiving operating characteristic (ROC) curves. For future synergic exploitation of multiple SAR missions, the developed dataset, composed of Sentinel-1, SAOCOM, and COSMO-SkyMed images, is comprehensive, having images gathered over the same area with a short time lag (below 15 min). Finally, the diversified processing chains and the results for each mission product and scenario are discussed. Being the first dataset of single-look complex (SLC) SAR multi-frequency data, the present work intends to encourage additional investigation in this promising field of research.
This content is subject to copyright.
Citation: Del Prete, R.; Graziano,
M.D.; Renga, A. Unified Framework
for Ship Detection in Multi-
Frequency SAR Images: A
Demonstration with COSMO-
SkyMed, Sentinel-1, and SAOCOM
Data. Remote Sens. 2023,15, 1582.
https://doi.org/10.3390/rs15061582
Academic Editors: Weimin Huang,
Deepak R. Mishra and Ana C. Brito
Received: 7 February 2023
Revised: 7 March 2023
Accepted: 10 March 2023
Published: 14 March 2023
Copyright: © 2023 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
remote sensing
Article
Unified Framework for Ship Detection in Multi-Frequency SAR
Images: A Demonstration with COSMO-SkyMed, Sentinel-1,
and SAOCOM Data
Roberto Del Prete * , Maria Daniela Graziano and Alfredo Renga
Department of Industrial Engineering, University of Naples Federico II, P.le Tecchio 80, 80125 Naples, Italy
*Correspondence: roberto.delprete@unina.it; Tel.: +39-081-7682160
Abstract:
In the framework of maritime surveillance, vessel detection techniques based on space-
borne synthetic aperture radar (SAR) images have promoted extensive applications for the effective
understanding of unlawful activities at sea. This paper deals with this topic, presenting a novel
approach that exploits a cascade application of a pre-screening algorithm and a discrimination phase.
Pre-screening is based on a constant false alarm rate (CFAR) detector, whereas discrimination exploits
sub-look analysis (SLA). For the first time, the method has been validated with experiments on
multi-frequency (C-, X-, and L-band) SAR images, demonstrating a significant reduction of up to
40% in false alarms within highly congested scenarios, along with a notable enhancement of the
receiving operating characteristic (ROC) curves. For future synergic exploitation of multiple SAR
missions, the developed dataset, composed of Sentinel-1, SAOCOM, and COSMO-SkyMed images, is
comprehensive, having images gathered over the same area with a short time lag (below 15 min).
Finally, the diversified processing chains and the results for each mission product and scenario are
discussed. Being the first dataset of single-look complex (SLC) SAR multi-frequency data, the present
work intends to encourage additional investigation in this promising field of research.
Keywords:
syntheticaperture radar; maritime monitoring; multi-frequency; multi-mission; ship
detection; CFAR; sublook analysis
1. Introduction
SAR data possess the potential to monitor various ocean surface features, such as
ocean surface turbulence [1], including the detection of ship wakes [25]. Therefore, there
is considerable interest in automatic ship detection in SAR images [
6
], which can provide
quantifiable measurements of physical properties such as length and shape as well as
dynamic movements such as speed and bearing. Notably, the detection of vessels can be
attained both with airborne [7], spaceborne [8] and in situ instrumentation [9].
Concerning the latter, the Automatic Identification System (AIS), a very high-frequency
(VHF) transceiver built originally for collision avoidance, is the most massive source of
information used for maritime monitoring by Vessel Traffic Services (VTS) [
10
]. The
broadcasted messages contain useful information about a vessel’s identity, position, speed,
course, destination, and other data that are critical for maritime control and navigational
safety [
11
]. This critical information is delivered in a ship-to-ship and in a ship-to-shore
fashion with transmissions to AIS stations (Figure 1).
The initial worries regarding the efficacy of an AIS-based monitoring system can
emerge when examining its coverage at sea, which is up to 20 nautical miles without
repeaters. However, the major issue with the AIS lies in its “cooperative” nature. The
transmitter can be purposefully set off during unlawful activities. In such a circumstance,
the ship becomes a “dark vessel” (i.e., a vessel that operates without an AIS transponder
or with it turned off [
12
]). Still, the AIS legislation retains its carrying obligation only
for certain classes of vessels. As stated in the SOLAS regulation [
13
], all passenger ships
Remote Sens. 2023,15, 1582. https://doi.org/10.3390/rs15061582 https://www.mdpi.com/journal/remotesensing
Remote Sens. 2023,15, 1582 2 of 17
(regardless of size), international voyaging ships of a gross tonnage (GT) of 300 or more,
and 500 GT and greater cargo ships not embarked on international journeys are obliged
by the International Maritime Organization (IMO) to be equipped with AIS [
13
]. Finally,
situations of corrupted or incorrect AIS messages are prone to occur [
14
]. Therefore, while
definitely contributing to maritime domain awareness (MDA), AIS information is unable
to render the entire maritime picture. To actually be helpful, the AIS messages must
be used in cooperation with other sensors, specifically non-cooperative ones that also
show wider coverage [
15
]. Mainly for these reasons, satellite technologies are currently
integrated into marine surveillance services and procedures because they provide cost-
effective remote monitoring, a worldwide scope, regular updates, and a large volume of
data gathered [
11
,
16
,
17
]. Even if optical imagers started gaining attention [
16
], spaceborne
synthetic aperture radars (SARs) remain the most preferred choice because they offer unique
characteristics that make them particularly tailored to supporting AIS-based monitoring
systems. The SAR imaging mechanism exploits microwave pulses interacting with a target
and returning to the sensor. Ships are usually man-made targets, making their backscattered
energy significantly higher than the surrounding clutter. Therefore, a vessel usually appears
as a cluster of bright pixels in an SAR image with few features that are identifiable. Notably,
SAR is an active sensor not facing the disadvantage of operating only during the daytime.
It is worth noting that most illegal activities take place at night. Moreover, the transmitted
electromagnetic wave in the typical range of utilization (1–10 GHz) is not significantly
affected by cloud cover or precipitation, thus making the imaging system able to penetrate
clouds and detect vessels even during nighttime [18].
Figure 1. Pictorial view of a spaceborne SAR gathering an image over a coastal area.
For achieving effective maritime surveillance, not only the utilization of a non-cooperative
approach but also the synergic exploitation of multi-frequency/multi-mission (MM/MF) data
is essential. This is to take advantage of higher revisit times. This aim is approached by the
present work through proposing a custom algorithm for ship detection adapted to three
different SAR missions: Sentinel-1, SAOCOM, and COSMO-SkyMed. The algorithm uses
the fast and efficient constant false alarm rate (CFAR) [
19
22
] together with the sub-look
analysis (SLA) [
23
26
] discrimination technique. There is a wide corpus of research dealing
with ship detection in SAR images, and the detection techniques in SAR imagery are
influenced by several different key parameters, but the research on SAR ship detection
can be divided into categories based on the physical property exploited. The backscatter-
based methods [
27
,
28
] utilize the radar cross-sections (RCSs) [
18
] of the vessels. They
Remote Sens. 2023,15, 1582 3 of 17
are fast and easy from a design point of view but have low performance, since they are
typically affected by false alarms and ambiguities [
24
26
,
29
]. Polarization-based [
26
,
30
34
] approaches leverage the polarimetric scattering mechanism to separate ships from
clutter. These approaches are generally more robust but are usually time-consuming and
computationally intensive. That aside, for polarimetric scattering decomposition [
35
],
quad-pol SAR imagery is required. The geometry-based methods [
36
,
37
] search for specific
geometric features, such as the length, width, aspect ratio, perimeter, area, or contour. They
demand an adequate template library and high-resolution SAR imagery. Feature-based
methods use local feature descriptors (e.g., histograms of oriented gradients (HOG) [
38
],
scale-invariant feature transform (SIFT) [
39
], and Haar-like features [
40
]). The methods
show maturity in feature design, but they are time-consuming and weak in migration.
Very recently, thanks to the large availability of Earth observation data, deep learning
methods [
41
47
] were also introduced in the ship detection community. These techniques
learn the non-hand-engineered abstract features from large annotated data for extrapolating
specific patterns during inference time. Promising performance has been demonstrated
even near coasts and reefs without the need for land separation [48]. The disadvantage of
these methods is in the supervised learning approach that demands large labeled datasets.
As a further relevant aspect, it is important to acknowledge that the presence of sea
clutter can have a significant impact on the detection of vessels in SAR imagery. The radar
signal reflected by the sea surface is susceptible to scattering caused by minor fluctuations
on the sea’s surface, generating noise that can compromise the accuracy of statistical
computations and produce false detections [
49
]. Various factors, including wind-generated
waves, currents, and surface roughness, can trigger sea clutter, the manifestation of which
is contingent on the SAR frequency and polarization, as well as the incidence angle and
resolution of the sensor. To address the challenge of sea clutter in SAR imagery, the
SLA approach can be employed to analyze pixel stability across the frequency spectrum.
By using this approach, the effects of sea clutter can be mitigated, and the detection
performance of vessels can be enhanced [23].
In the framework of the COastal Area monitoring with SAR data and multimis-
sion/multifrequency Techniques (COAST) project, funded by the Italian Space Agency
(ASI), a novel dataset has been developed utilizing MM/MF imagery. The comprehensive
dataset enables the testing of the effectiveness of several missions under comparable cir-
cumstances. To the knowledge of the authors, this constitutes the first single-look complex
(SLC) MM/MF SAR dataset, and this is also a major novelty of this work. Another inno-
vation lies in its specific attention to in-shore areas, which are typically characterized by
perturbing phenomena affecting the detection performance. The latter can include the fast
dynamics of vessels’ motion near ports, ambiguities generated by land-strong scatterers, or
saturation or anomalous side lobe pattern effects.
This manuscript is structured as follows. Section 2details the development of the
MM/MF SAR dataset and its ancillary information. Section 3details the methodology
adopted in this work from the pre-processing of the data to the implementation of a novel
detector. Then, Section 4discusses the results achieved by means of a large-scale validation
approach on the proposed dataset. Finally, Section 5draws the conclusions of the present
study while also pointing out new future directions.
2. Multi-Mission/Multi-Frequency SAR Dataset
As stated above, for the construction of the MM/MF dataset, three missions were
considered: the Italian COSMO-SkyMed (X-Band), the Argentinian SAOCOM (L-Band),
and the European Sentinel-1 (C-Band). The characteristics of the selected products are
briefly described in Table 1, which demonstrates how the images differ beyond the work-
ing frequency band. Concerning the Sentinel-1 images, they were gathered in dual-pol
mode (VV + VH) or interferometric wide (IW) swath mode, even if VH-only data were
processed. COSMO-SkyMed and SAOCOM were instead acquired in stripmap HH and
VH polarization, respectively.
Remote Sens. 2023,15, 1582 4 of 17
Table 1.
MM/MF product specifications in terms of acquisition mode, pixel spacing, and polarization.
Mission Acquisition Mode Pixel Spacing
(range ×azi) (m) Polarization Swath
(km)
COSMO-SkyMed StripMap 0.5 ×0.5 HH 40
Sentinel-1
IW (Interferometric Wide Swath)
2.3 ×13.9 VH 250
SAOCOM StripMap <10 ×10 VH 65
2.1. Selected Scenarios
Data collection started from the selection and identification of the region of interest.
For the present study, three different scenarios were selected for the purpose of maritime
monitoring: Adriatic Sea, Sardinia, and the Egadi Islands (Figure 2).
b
a
c
Figure 2.
Selected scenarios of interest for the realization of the MM/MF dataset: (a) Adriatic Sea,
(b) Sardinia, and (c) the Egadi Islands.
It must be emphasized once more that the goal was to maximize the benefits of
MM/MF SAR images for maritime surveillance. As a result, a short time gap between suc-
cessive acquisitions over the same region was required. A tool that inspected the effective
spatial and temporal couplings of the MM/MF products was developed specifically for this
purpose. The tool provides highlights of the various acquisition availabilities, reporting
useful insights about the three SAR missions under consideration. Finally, the value of the
project’s supplementary AIS data was explored.
The Egadi Islands Marine Reserve, with its 53,992 hectares, is not only Europe’s largest,
but it also has the peculiarity of being the initial point of arrival for several marine species
whose migrations are frequently linked to the flow of the Atlantic current. In recent years,
there have been several complaints about illegal activities in the Egadi Islands’ marine
protected area, such as trawling in shallow waters or the use of illegal nets longer than the
2.5 km required by law for underwater fishing in prohibited areas [
50
]. Sardinia’s scenario
concerns violations related to the “waste cycle” and polluting discharge. In recent years,
there have also been several seizures of drugs and weapons in Sardinian ports and waters.
Additionally, according to recent research [
51
,
52
], Sardinia has the absolute record of seized
fish products. Finally, in the Adriatic Sea, in general, there has been a drastic decline in fish
Remote Sens. 2023,15, 1582 5 of 17
stocks due to intensive fishing, which has profoundly changed the marine environment.
This is an area where excessive trawling has had a very strong impact, so much so that fish
stocks in the Adriatic Sea have been greatly reduced [
53
]. Illegal fishing is not the only
unlawful activity, as arms and drug trafficking and smuggling are also widespread.
2.2. Footprint Matching
The footprint matching algorithm (Algorithm 1) is a novel contribution of this paper
that oversees coupling different SAR products on the same spot and proceeds in two
cascaded steps. In the first stage, SAR products are filtered on a temporal basis, considering
two products matched if their sensing period difference is below 15 min. This key value
was determined with a heuristic rationale, whereby ships after that time difference could
not be matched between the images. The algorithm serves as a pre-screening step and
is followed by spatial matching to filter the product that covers the same area of interest
(AOI). The spatial matching is carried out by exploiting the footprint data contained in
the product metadata. The area of intersection is used to establish a correct spatial match
between two footprints. Finally, to conclude the analysis, a visual inspection is performed.
The footprints are plotted in interactive *.html maps, and the quality of the intersection is
evaluated both in terms of area covered and area of interest, verifying that the intersections
cover sea zones.
Algorithm 1: Footprint matching MM/MF SAR products.
Input: MM/MF Products
for each product pido
for each product pjwith j>ido
if |titj|<15 then
if piand pjcover the same area of interest then
matchedProducts matchedProducts (pi,pj);
return matchedProducts;
An example of a map realized for the scenario of Egadi Island is reported in Figure 3,
with which it is possible to observe the Sentinel-1 and COSMO-SkyMed coupling.
Figure 3.
Sentinel-1 (Prod ID: S1B_W_SLC_1SDV_20210316T170404_20210316T170431_026043_
031B74_81C7) and COSMO-SkyMed (Prod ID: 1762255) product pairing map in the Egadi Islands scenario.
Table 2shows the number of couplings in each scenario selected for the COAST
project, detailing separately the various SAR mission parings. Notably, in the Egadi Islands
Remote Sens. 2023,15, 1582 6 of 17
region, only the COSMO-SkyMed and Sentinel-1 products found matches. For major details
regarding the coupled products, their relative geometries, orbit types, and platforms can be
found in [54].
Table 2. Number of couplings for each scenario of interest of the MM/MF dataset.
Pairing
Region Adriatic Sea Egadi Islands Sardinia
COSMO-SkyMed and Sentinel-1 15 32 55
COSMO-SkyMed and SAOCOM 5 NA 23
Sentinel-1 and SAOCOM 12 NA 10
2.3. AIS Data
Ancillary AIS data are a useful resource included in the dataset to provide information
on specific targets. The AIS messages can be categorized into three types: static, dynamic,
and voyage-related information. Only the dynamic one was updated frequently (<10 min),
with a rate changing according to the vessel speed and course variation, ranging from
a few seconds for very fast ships to several minutes for slow or moored ships. When
collaborative ships were available, their position, contained in the SAR products, was
stored in a database. Nonetheless, the storage policy of [
55
] saves the messages with a
temporal resolution of around 1 min. Therefore, data were pre-processed beforehand, with
Hermite interpolation taking as a reference time the central time of acquisition of each
SAR image. Aside from that, when an insufficient number of points made it infeasible to
execute an interpolation, an extrapolation technique was used. However, in many cases,
this approach resulted in wrong placement of the ship location. This misleading result was
caused by wrong extrapolation, erroneous AIS messages, or sudden route changes of the
ships. Hence, the actual positions of ships were collected by visually inspecting each SAR
image and manually labeling each vessel.
3. Method
The targets that are not truthful vessels acting as powerful signature ghosts are labeled
as ambiguities in the SAR literature [
23
]. The method developed in this manuscript aims
at mainly filtering the latter. The causes of ambiguities may be traced back to the limited
sampling of the SAR pulses, which affects the Doppler spectrum [
25
]. Generally, SAR
ambiguities can be produced in two ways. In the first case, the ambiguities are brought on
by large ships. In this scenario, the ship’s brightness creates ambiguity that may be stronger
than the surrounding clutter, which might cause ghosts in the SAR picture to resemble
the ship’s actual signature. In the second case, the ambiguities are caused by land targets.
Due to the low backscattering from the water in this scenario, which is a characteristic of
coastal zones, the ambiguities produced by land targets are shown as bright targets over
the sea’s surface. In the case of high-resolution SAR images, these artifacts become more
prominent. In essence, as the spatial resolution grows, so does the compression gain in
SAR image formation. As a result, the improvement of the dynamic range of SAR images
increases the intensity of strong point scatterers. The following sections describe in detail
the pre-processing chains for each SAR product and the CFAR+SLA detector.
3.1. Pre-Processing Chains
A general understanding of the pre-processing chains implemented for the detection
of visible ships in the MM/MF products is illustrated in Table 3, where each pre-processing
step applied to the different SAR products is detailed with a checkmark. Sea-land seg-
mentation is a mandatory step that can greatly minimize false alarm rates and enhance
follow-up processing efficiency.
Remote Sens. 2023,15, 1582 7 of 17
Table 3. Pre-processing operators used for each SAR product.
Operator
Product COSMO-SkyMed SAOCOM Sentinel-1
Multilook X
Thermal noise removal X
TOPSAR deburst X X
Land masking X X X
Calibration X X
As also exhibited in Table 3, apart from land masking (LM) processing, SAOCOM
products do not require any further dispensation. The multilook operator, applied only
on COSMO-SkyMed products, is not mandatory but suggested to reduce the computa-
tional burden. Significant concerns regard the LM operator, which usually requires an
accurate model of the coastline, whereas other researchers prefer to use specific extraction
techniques [56]. In this paper, the land polygons were extracted from [57].
3.2. The CFAR+SLA Detector
The methodology developed proceeds in two steps in cascade: pre-screening and
discrimination. First, the target proposals are generated with a traditional CFAR algorithm,
and then these ones are discriminated with a spectrum analysis technique (i.e., sub-look
analysis (SLA)). The adaptive threshold applied uses the nested windows approach, in
which there are three windows around each pixel under testing: a target window (TW),
a guard window (GW), and a background window (BW). The adaptive threshold is the
basis of the pre-screening process. The size of the target window should be approximately
equal to the size of the smallest object to detect, the size of the guard window should be
approximately equal to the size of the largest object, and the size of the background window
should be sufficiently large to estimate the local background statistics accurately. Indeed,
the algorithm leverages the statistical modeling of the background clutter [
58
61
], with
which a probability density function
fpd f (x)
can be associated. Thus, the design parameter
Tcan be computed by the from user-selected PFA as follows:
PFA =1
T
Z
fpd f (x)dx =
Z
T
fpd f (x)dx. (1)
It is worth highlighting that the Tthreshold is established empirically during practical
implementation, with formulas solely serving as a directional guide. This is because a
sufficiently accurate but also general enough model for the sea background is not typically
available. Once the prompted background mean
µb
and standard deviation
σb
use pixels in
the background ring and the mean value
µt
of the target window, a region is a potential
target candidate if
µt>µb+σbT. (2)
Based on the output of Equation (2), the algorithm first groups the contiguous detected
pixels into a single cluster and then extracts the width and length information from the
clusters. Finally, clusters that are too large or too small are excluded based on these
measurements and user input discrimination criteria. This first level of filtering is called
geometric discrimination. In conclusion, CFAR detection is designed to search for pixels
that are unexpectedly bright in comparison with those in the surrounding sea, although
SAR ambiguities or sea clutter may also fit this criterion. The pseudo-code of the CFAR
algorithm is reported in Algorithm 2.
Remote Sens. 2023,15, 1582 8 of 17
Algorithm 2: CFAR algorithm.
Input: data, background, guard, and target window size, threshold T
Step 0: Raster Tiling input data (1px stride);
for each tile do
1: Using the nested windows: background, guard, and target window:
1.1 Calculate the average value of the background window;
1.2 Calculate the standard deviation of the background window;
1.3 Calculate the average value of the target window;
2: Use Equation (2):
if True
then Targets Targets newTarget;
3: Cluster continuous pixels marked as target;
3.1 Apply Geometric Discrimination;
return Targets
To tackle the false alarm issue, the second level of discrimination employs sub-look
analysis of the selected region of interest for the removal of false alarms. The physical
rationale is that the Doppler spectra of ambiguities and targets are distinguished. As
shown in Figure 4, subsets and metadata are fed as input to the SLA algorithm. The typical
metadata information required by the sub-look processor contains parameters such as the
pulse repetition frequency (PRF) or the bandwidth of processing in the azimuth [23].
6$5352'8&7'HWHFWLRQV0HWDGDWD;0/6HQWLQHO&)$56/$&26026N\0HG6$2&206$2&200XOWLORRN7KHUPDO1RLVH5HPRYDO7236$5'HEXUVW/DQG0DVNLQJ*HRPHWULF'LVFULPDWLRQ6XEVHWWLQJ&DOLEUDWLRQ/DQG0DVNLQJ/DQG0DVNLQJ&DOLEUDWLRQ
Figure 4. Complete processing flow of the CFAR and SLA algorithms.
Sub-looks are generated starting from the SLC regions of interests detected in the
previous steps. As detailed in [
23
], one-dimensional sub-look generation is conceived,
which is the most common approach for ship detection. Therefore, sub-looks can be gen-
erated either in the azimuth or in a range. It is worth recalling that there is no significant
difference in ship detection performance between the range and azimuth sub-looks even
when moving ships are imaged, but azimuth sub-looks are used in this work for ambiguity
rejection. The number of sub-looks
NSL
, the bandwidth of the sub-looks
BSL
, and the
frequency separation between the centroids of two close looks
fc
constitute the relevant
parameters that must be set. In general,
BSL
is equally configured for all sub-looks. This
ensures dealing with sub-looks having the same resolution and thus allowing a fair com-
parison between them. Indeed, the fraction of
BSL
covered by a sub-look is an index of
the degradation of the resolution with reference to the original resolution of the SLC data.
Regarding
NSL
, for ship detection purposes, usually only two sub-looks are considered.
However, for estimating the incoherent entropy (IE) [
23
], a minimum of three sub-looks
was used. Concerning the location of each sub-look, the common approach is to consider
the sub-looks as equally spaced along the available bandwidth. It is worth remembering
that sub-looks are overlapped in frequency if
fc<BSL
. Table 4displays the parameters
used for the sub-look analysis of Sentinel-1, COSMO-SkyMed, and SAOCOM missions.
Remote Sens. 2023,15, 1582 9 of 17
Table 4.
Parametersused for the sub-look analysis of Sentinel-1, COSMO-SkyMed, and SAOCOM
missions.
Parameter
Mission Sentinel-1 COSMO-SkyMed SAOCOM
BSL 102.0 Hz 466.6 Hz 372.0 Hz
fc102.0 Hz 466.6 Hz 372.0 Hz
Wr,Wc7, 17 17, 17 3, 17
After sub-look generation, the IE relative to each region of interest was prompted.
In the end, a threshold algorithm was applied by considering stable pixels in a nested
window fashion. In fact, considering a matrix of IE prompted for a subset, a statistical
process analyzed the average values inside and outside the small 30
×
30 pixel region at the
center of each subset (Figure 5). As for the CFAR, a buffer window of 70
×
70 pixels was
considered to reduce disturbances in the computations. To summarize, the discrimination
algorithm output a class for the target or ambiguity depending on the stability of the pixels
of the target, with reference to the pixel stability of the background. In conclusion, the
CFAR+SLA detector can be considered an extension of [
23
], where the IE pertained only to
the pre-screened targets and the thresholds were prompted in a nested fashion.
Figure 5.
Visual representation of IE generation from three sub-looks (
left
) and its calculation for a
Sentinel-1 product (right), in which a target and its ambiguity are highlighted.
Some demonstrative examples of IE for the COSMO-SkyMed, Sentinel-1, and SAO-
COM ship targets and ambiguities are reported in Figures 68, respectively. These samples
are very useful for showing the crucial behaviors of IE under different frequency bands
while still not damaging the effectiveness of the discriminator.
Figure 6.
Example of IE calculated for ship target (
left
) and ambiguity (
right
) for a CSK product.
(Product ID: CSKS1_SCS_B_HI_05_ H H_R D_SF_20201017164505_20201017164512).
Remote Sens. 2023,15, 1582 10 of 17
Figure 7.
Example of IE calculated for ship target (
left
) and ambiguity (
right
) for a Sentinel-1 product.
(Product ID: S1B_IW_SLC__1SDV_20200510T164813_20200510T164840_021522_028DCF_60B1).
Figure 8.
Example of IE calculated for ship target (
left
) and ambiguity (
right
) for a SAOCOM product.
(Product ID: S1A_OPER_SAR_EOSSP__CORE_L1A_OLVF_20210122T210006).
4. Experimental Analysis
4.1. Performance Indicators
To monitor the overall performance of the detection algorithm, the detection probabil-
ity, false alarm rate, and consequently the receiver operating characteristic (ROC) curves
were defined. It should be noted that in the literature, the metrics of the detection probabil-
ity
Pd
and false alarm probability
Pf
are defined differently from author to author [
62
]. For
example, some authors prefer to relate the false alarm rate to the area of the observation
scenario [27]. In this study, such metrics were adopted as follows:
Pd=Ndt
Nt
Pf=NDet Ndt
NDet
(3)
having denoted
Ndt
as the number of target-coupled detections,
Nt
as the number of
targets in the scene, and
NDet
as the total number of detection algorithm outputs. Once the
metrics are defined, it is possible to measure the relative importance of the CFAR algorithm
parameters for the detection performance. It is noteworthy that, having defined the metrics
in such a way, the probability of detection and false alarm rate results are decoupled.
However, before calculating these curves, an attempt was made to solve one of the issues
of this adaptive threshold approach. While the CFAR algorithm is well-established and
widely accepted in the scientific community, it is not without its faults. The multiple
detections of the same target represent one of the most typical examples. To address this
issue and consequently minimize the false alarm rate, a detection suppression technique
was developed, combining nearby detections that are located less than 150 m apart from
each other. The new location is supposed to be placed in the middle of the two. Notably,
the threshold was experimentally determined.
4.2. Local Analysis
A study of the CFAR+SLA algorithm’s performance on a COSMO-SkyMed product
(Figure 9) is presented here.
This scene was chosen for its unique characteristics since it represents a very difficult
case with intricate coasts, azimuth ambiguities, and ships in close proximity to one another
and the coast. In fact, the image includes the port of Taranto, which features many potential
sources of false alarms due to its geographical configuration. The presence of docks and
other metal buildings is a further source of uncertainty. Finally, there is an artifact in the
image, specifically a bright stripe, which is most likely the result of a radio signal emitted
Remote Sens. 2023,15, 1582 11 of 17
by one of the ships in port and captured by the X-band satellite. The product was tested
with a low threshold value, background, guard, and target window sizes as given in Table 5.
In particular, the threshold value was computed by taking the value provided in Table 5
and converting it into a decimal fraction using the negative exponent of 10.
Figure 9.
Test case of the CFAR+SLA algorithm over the port of Taranto. Highlighted detail: a
probable radio signal transmitted from a ship and captured by X-band spaceborne SAR. Product ID:
CSKS
1
_SCS_B H I_
05
_HH_RD_SF_
20200323164522
_
2020032316
_
4529, processed by University of
Naples Federico II under the COAST license of the Italian Space Agency (ASI). Original COSMO-
SkyMed Product ©ASI (2020).
Table 5. Configuration of the paramaters of the CFAR algorithm.
BW GW TW PFA (10x)Min Target Size Max Target Size
800 m 400 m 30 m 4.5 30 m 800 m
This is the default option for running the algorithm in order to increase the detection
probability while decreasing the false alarm rate. In any event, this permits the discrimina-
tion algorithm’s performance to be tested across a large number of detections. The overall
accuracy of the CFAR+SLA algorithm was evaluated through ship and ambiguity detection,
having defined the accuracy as the ratio between the true positives and true positives plus
false positives. Ultimately, pre- and post-application results of the discrimination chain
were evaluated in terms of the detection probability and false alarm rate. The results were
obtained and reported in Table 6.
Table 6.
Accuracy of the CFAR+SLA algorithm and performance in terms of probability of detection
and false alarm rate pre- and post-application of the discrimination pipeline.
Accuracy CFAR CFAR+SLA
Vessels 92.3% PdPfPdPf
Ambiguities 100%
Global 95.6% 100% 47.8% 100% 7.6%
Remote Sens. 2023,15, 1582 12 of 17
4.3. Global Analysis
This section details the performance analysis to estimate the improvement achieved
with the CFAR-SLA algorithm with respect to the application of CFAR only. For this
purpose, MM/MF products coupled by spatio-temporal matching were analyzed with
the CFAR-SLA algorithm in the different identified regions. In more detail, after labeling
the products individually, the performance was derived via ROC curves. What follows is
a brief description of the curves obtained in each reference scenario. Since the AIS data
were not fully usable, it should be noted that the goodness of the curves depended on the
annotator’s recognition skill. The ROC curves were calculated by varying the threshold
parameters of the detection algorithm and keeping the other parameters fixed, as in the
basic configuration shown in Table 5. Specifically, the threshold was linearly sampled in
the interval [4.5, 19.25].
4.3.1. Egadi Islands
In the Egadi Islands region, the performance of the algorithm is highlighted in Figure 10.
It should be taken into account that in this region, the land portions are far smaller than
the sea portions. This made it possible to reduce the false alarm rate generated by land
ambiguities. As can be seen from the results, the performances of Cosmo-SkyMed and
Sentinel-1 were remarkable. Specifically, measuring the performance as the area under the
curve (AUC), COSMO-SkyMed achieved a value of 0.91 with CFAR only, which increased to
0.95 after the the application of the discrimination pipeline. Concerning Sentinel-1 products,
the AUC improved by more than 15% before and after application of the discrimination
algorithm, going from 0.60 to 0.70.
Figure 10.
ROC evaluation of Cosmo-SkyMed (89 targets labeled) (
a
) and Sentinel-1 (290 targets
labeled) (b) products in the Egadi Islands region.
4.3.2. Sardinia
The evidence of Figure 11 shows how the performance achieved in the region of
Sardinia is noteworthy, proving the effectiveness of the discrimination algorithm.
Indeed, the COSMO-SkyMed products clearly demonstrate performance improve-
ments. As can be seen, albeit with a marginal loss in accuracy in the early part of the
graph, the COSMO-SkyMed products experienced a performance increase from the already
high AUC value of 0.88 to 0.91. By observing Figure 11b, the SAOCOM products showed
a small decay in the CFAR-SLA curve around the 0.7 value of Pf. However, as can be
appreciated, the AUC markedly increased from a value of 0.78 to 0.82. The Sentinel-1
products (Figure 11) demonstrated good performance, which increased again using the
sub-aperture algorithm. In fact, the performance enhancement increased the AUC from
0.72 to 0.77.
Remote Sens. 2023,15, 1582 13 of 17
Figure 11.
ROC evaluation of COSMO-SkyMed (121 targets labeled) (
a
), SAOCOM (36 targets labeled)
(b), and Sentinel-1 (333 targets labeled) (c) products in the Sardinia region.
4.3.3. Adriatic Sea
When analyzing the products in the Adriatic Sea, it is again clear from the results
obtained that there was an increase in performance after the removal of false alarms. The
relevant curves can be seen in the graphs in Figure 12. In contrast to the previous case, the
COSMO-SkyMed products showed a performance increase, especially in the early part of
the graph. The accuracies were also remarkable in this scenario, going from a value of 0.78
to 0.87 for the AUC. The same observations can also be repeated for the SAOCOM products
(Figure 12), showing an increased AUC after application of the discriminator algorithm. As
can be seen from the bottom graph in Figure 12, the AUC increased from a value of 0.73 to
0.77. Finally, despite slightly lower accuracy values at the end of the graph, the Sentinel-1
products showed an improvement in detection performance, as the area under the curve
improved from 0.83 to 0.88.
4.3.4. Area under the Curve
In essence, the global performance of the COSMO-SkyMed and Sentinel-1 as well
as the COSMO-SkyMed and SAOCOM pairs were evaluated and reported by means of
Figure 13. The latter reported the AUC before and after application of the sub-aperture
analysis algorithm, divided into the three scenarios of interest. As can be seen from the
graphs, the COSMO-SkyMed performance was remarkable in every scenario and condition,
especially in the Egadi Islands region, where large portions of the sea overlie the few areas
of land. This again testifies to the importance of the correct execution of land separation.
Remote Sens. 2023,15, 1582 14 of 17
Figure 12.
ROC evaluation of Cosmo-SkyMed (108 targets labeled) (
a
), SAOCOM (113 targets labeled)
(b), and Sentinel-1 (331 targets labeled) (c) products in the Adriatic region.
Figure 13.
Comparative analysis of improved performance in terms of AUC for the considered
sensors and the selected scenarios. CSK = COSMO-SkyMed, SAO = SAOCOM, and SEN = Sentinel-1.
5. Conclusions
Within the time frame of 18 months, the present study assessed the capability of a
cascade detector for ship detection purposes on multiple SAR frequency bands (L-, C-, and
X-bands). The processing chains and the constructed dataset of MM/MF SLC SAR products
constitute an important contribution of this work. The results attained have confirmed
the effectiveness of the developed approach, showing an increase in performance in terms
of improvement of the AUC and reduction of false alarms. Indeed, in a very congested
scenario, such as the port of Taranto, the reduction of the false alarm rate was estimated to
be about 40%.
Remote Sens. 2023,15, 1582 15 of 17
While undoubtedly recognizing the ghost targets, it must be pointed out that SLA
involves the solution of an eigenvalue problem which is computationally intensive, with an
increase in processing time that grows quadratically with the dimension of the considered
tile. It should be taken into account how the time demanded by the discrimination phase
matches the one from pre-selection. Therefore, to reduce the time required to process a
panoramic SAR product, further research will analyze the capabilities offered by a deep
learning-based technique on the developed dataset. Taking full advantage of the SAR
spectrum is definitely a path that must be exploited with artificial intelligence.
Author Contributions:
Conceptualization, A.R. and M.D.G.; methodology, R.D.P.; software, R.D.P.;
validation, M.D.G., A.R.; formal analysis, R.D.P.; investigation, R.D.P.; resources, A.R. and M.D.G.;
data curation, R.D.P.; writing—original draft preparation, R.D.P.; writing—review and editing, R.D.P.,
A.R. and M.D.G.; visualization, R.D.P.; supervision, A.R.; project administration, A.R.; funding
acquisition, M.D.G. All authors have read and agreed to the published version of the manuscript.
Funding:
This research was funded by the Italian Space Agency (ASI), grant number N.2021-11-U.0.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement:
Due to confidentiality concerns and privacy regulations, the data used
in this study cannot be distributed publicly.
Acknowledgments:
This work was developed in the framework of the Italian Space Agency’s “Study
of new methods and techniques based on the utilization of multimission/multifrequency SAR data”
project “COastal Area monitoring with SAR data and multimission/multifrequency Techniques -
COAST”, ASI Contract N. 2021-11-U.0.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Marghany, M. Nonlinear Ocean Dynamics: Synthetic Aperture Radar; Elsevier: Amsterdam, The Netherlands, 2021.
2.
Graziano, M.D.; D’Errico, M.; Rufino, G. Wake component detection in X-band SAR images for ship heading and velocity
estimation. Remote Sens. 2016,8, 498. [CrossRef]
3.
Graziano, M.D.; D’Errico, M.; Rufino, G. Ship heading and velocity analysis by wake detection in SAR images. Acta Astronaut.
2016,128, 72–82. [CrossRef]
4.
Graziano, M.D. Preliminary results of ship detection technique by wake pattern recognition in SAR images. Remote Sens.
2020
,
12, 2869. [CrossRef]
5.
Del Prete, R.; Graziano, M.D.; Renga, A. First results on wake detection in SAR images by deep learning. Remote Sens.
2021
,
13, 4573. [CrossRef]
6.
Vachon, P.W.; Campbell, J.; Bjerkelund, C.; Dobson, F.; Rey, M. Ship detection by the RADARSAT SAR: Validation of detection
model predictions. Can. J. Remote Sens. 1997,23, 48–59. [CrossRef]
7.
Joshi, S.K.; Baumgartner, S.V. Range-Doppler tracking of ships using single-channel airborne radar data. In Proceedings of the
EUSAR 2021; 13th European Conference on Synthetic Aperture Radar, VDE Conference, online, 29 March–1 April 2021; pp. 1–6.
8.
Eldhuset, K. An automatic ship and ship wake detection system for spaceborne SAR images in coastal regions. IEEE Trans. Geosci.
Remote Sens. 1996,34, 1010–1019. [CrossRef]
9.
Shao, Z.; Wu, W.; Wang, Z.; Du, W.; Li, C. Seaships: A large-scale precisely annotated dataset for ship detection. IEEE Trans.
Multimed. 2018,20, 2593–2604. [CrossRef]
10. Young, W. What are vessel traffic services, and what can they really do? Navigation 1994,41, 31–56. [CrossRef]
11.
Goudossis, A.; Katsikas, S.K. Towards a secure automatic identification system (AIS). J. Mar. Sci. Technol.
2019
,24, 410–423. [CrossRef]
12.
Tetreault, B. Use of the Automatic Identification System (AIS) for maritime domain awareness (MDA). In Proceedings of the
OCEANS 2005 MTS/IEEE, Washington, DC, USA, 19–23 September 2005; Volume 2, pp. 1590–1594. [CrossRef]
13.
International Maritime Organization. “AIS transponders”, International Maritime Organization (2017) [Online]. 2019. Available
online: https://www.imo.org/en/OurWork/Safety/Pages/AIS.aspx (accessed on 7 December 2022).
14.
Zhang, D.; Li, J.; Wu, Q.; Liu, X.; Chu, X.; He, W. Enhance the AIS data availability by screening and interpolation. In Proceedings
of the 2017 4th International Conference on Transportation Information and Safety (ICTIS), Banff, AB, Canada, 8–10 August 2017;
pp. 981–986.
15.
Graziano, M.D.; Renga, A.; Moccia, A. Integration of Automatic Identification System (AIS) Data and Single-Channel Synthetic
Aperture Radar (SAR) Images by SAR-Based Ship Velocity Estimation for Maritime Situational Awareness. Remote Sens.
2019
,11,
2196. [CrossRef]
Remote Sens. 2023,15, 1582 16 of 17
16.
Kanjir, U.; Greidanus, H.; Oštir, K. Vessel detection and classification from spaceborne optical images: A literature survey. Remote
Sens. Environ. 2018,207, 1–26. [CrossRef] [PubMed]
17.
Iceye. Dark Vessel Detection for Maritime Security with SAR Data, [Online]. 2021. Available online: https://www.iceye.com/
use-cases/security/dark-vessel-detection/interactive- demo (accessed on 8 March 2023).
18. Curlander, J.C.; McDonough, R.N. Synthetic Aperture Radar; Wiley: New York, NY, USA, 1991; Volume 11.
19.
Wackerman, C.C.; Friedman, K.S.; Pichel, W.G.; Clemente-Colón, P.; Li, X. Automatic detection of ships in RADARSAT-1 SAR
imagery. Can. J. Remote Sens. 2001,27, 568–577. [CrossRef]
20.
Friedman, K.; Wackerman, C.; Funk, F.; Rowell, K.; Pichel, W.; Clemente-Colón, P.; Li, X. Validation of an automatic vessel
detection algorithm using SAR data and known vessel fleet distributions. In Proceedings of the IGARSS 2000. IEEE 2000
International Geoscience and Remote Sensing Symposium. Taking the Pulse of the Planet: The Role of Remote Sensing in
Managing the Environment. Proceedings (Cat. No. 00CH37120), Honolulu, HI, USA, 24–28 July 2000; IEEE: New York, NY, USA,
2000 ; Volume 5, pp. 2071–2073.
21.
Friedman, K.S.; Wackerman, C.; Funk, F.; Schwenzfeier, M.; Pichel, W.G.; Colon-Clemente, P.; Li, X. Analyzing the dependence
between RADARSAT-1 vessel detection and vessel heading using CFAR algorithm for use on fishery management. In Proceedings
of the Oceans 2003. Celebrating the Past... Teaming Toward the Future (IEEE Cat. No. 03CH37492), San Diego, CA, USA, 22–26
September 2003 ; Volume 5, pp. P2819–P2823.
22.
Friedman, K.S.; Wackerman, C.; Funk, F.; Pichel, W.G.; Clemente-Colón, P.; Li, X. Validation of a CFAR vessel detection algorithm
using known vessel locations. In Proceedings of the IGARSS 2001. Scanning the Present and Resolving the Future, Proceedings.
IEEE 2001 International Geoscience and Remote Sensing Symposium (Cat. No. 01CH37217), Sydney, Australia, 9–13 July 2001;
Volume 4, pp. 1804–1806.
23.
Renga, A.; Graziano, M.D.; Moccia, A. Segmentation of marine SAR images by sublook analysis and application to sea traffic
monitoring. IEEE Trans. Geosci. Remote Sens. 2018,57, 1463–1477. [CrossRef]
24.
Guarnieri, A.M. Adaptive removal of azimuth ambiguities in SAR images. IEEE Trans. Geosci. Remote Sens.
2005
,43, 625–633.
[CrossRef]
25.
Choi, J.H.; Won, J.S. Efficient SAR Azimuth Ambiguity Reduction in Coastal Waters Using a Simple Rotation Matrix: The Case
Study of the Northern Coast of Jeju Island. Remote Sens. 2021,13, 4865. [CrossRef]
26.
Velotto, D.; Soccorsi, M.; Lehner, S. Azimuth ambiguities removal for ship detection using full polarimetric X-band SAR data.
IEEE Trans. Geosci. Remote Sens. 2013,52, 76–88. [CrossRef]
27.
Pelich, R.; Longépé, N.; Mercier, G.; Hajduch, G.; Garello, R. AIS-based evaluation of target detectors and SAR sensors
characteristics for maritime surveillance. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2014,8, 3892–3901. [CrossRef]
28.
Wang, C.; Bi, F.; Zhang, W.; Chen, L. An intensity-space domain CFAR method for ship detection in HR SAR images. IEEE Geosci.
Remote Sens. Lett. 2017,14, 529–533. [CrossRef]
29.
Raney, R.K.; Princz, G.J. Reconsideration of azimuth ambiguities in SAR. IEEE Trans. Geosci. Remote Sens.
1987
,GE-25, 783–787.
[CrossRef]
30.
Touzi, R.; Hurley, J.; Vachon, P.W. Optimization of the degree of polarization for enhanced ship detection using polarimetric
RADARSAT-2. IEEE Trans. Geosci. Remote Sens. 2015,53, 5403–5424. [CrossRef]
31.
Touzi, R. On the use of polarimetric SAR data for ship detection. In Proceedings of the IEEE 1999 International Geoscience
and Remote Sensing Symposium. IGARSS’99 (Cat. No. 99CH36293), Hamburg, Germany, 28 June–2 July 1999 ; Volume 2,
pp. 812–814.
32.
Brekke, C.; Anfinsen, S.N. Ship detection in ice-infested waters based on dual-polarization SAR imagery. IEEE Geosci. Remote
Sens. Lett. 2010,8, 391–395. [CrossRef]
33.
Marino, A.; Sanjuan-Ferrer, M.J.; Hajnsek, I.; Ouchi, K. Ship detection with spectral analysis of synthetic aperture radar: A
comparison of new and well-known algorithms. Remote Sens. 2015,7, 5416–5439. [CrossRef]
34.
Wang, W.; Ji, Y.; Lin, X. A novel fusion-based ship detection method from Pol-SAR images. Sensors
2015
,15, 25072–25089.
[CrossRef]
35.
Xie, Q.; Ballester-Berman, J.D.; Lopez-Sanchez, J.M.; Zhu, J.; Wang, C. Quantitative analysis of polarimetric model-based
decomposition methods. Remote Sens. 2016,8, 977. [CrossRef]
36.
Zhu, J.; Qiu, X.; Pan, Z.; Zhang, Y.; Lei, B. Projection shape template-based ship target recognition in TerraSAR-X images. IEEE
Geosci. Remote Sens. Lett. 2016,14, 222–226. [CrossRef]
37.
Wang, C.; Bi, F.; Chen, L.; Chen, J. A novel threshold template algorithm for ship detection in high-resolution SAR images. In
Proceedings of the 2016 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Beijing, China, 10–15 July
2016 ; pp. 100–103.
38.
Song, S.; Xu, B.; Yang, J. SAR target recognition via supervised discriminative dictionary learning and sparse representation of
the SAR-HOG feature. Remote Sens. 2016,8, 683. [CrossRef]
39.
Zhou, D.; Zeng, L.; Zhang, K. A novel SAR target detection algorithm via multi-scale SIFT features. J. Northwest. Polytech. Univ.
2015,33, 867–873.
40.
Ai, J.; Tian, R.; Luo, Q.; Jin, J.; Tang, B. Multi-scale rotation-invariant Haar-like feature integrated CNN-based ship detection
algorithm of multiple-target environment in SAR imagery. IEEE Trans. Geosci. Remote Sens. 2019,57, 10070–10087. [CrossRef]
Remote Sens. 2023,15, 1582 17 of 17
41.
Zhang, T.; Zhang, X. High-speed ship detection in SAR images based on a grid convolutional neural network. Remote Sens.
2019
,
11, 1206. [CrossRef]
42.
Chen, C.; He, C.; Hu, C.; Pei, H.; Jiao, L. A deep neural network based on an attention mechanism for SAR ship detection in
multiscale and complex scenarios. IEEE Access 2019,7, 104848–104863. [CrossRef]
43.
Zhang, T.; Zhang, X.; Shi, J.; Wei, S. Depthwise separable convolution neural network for high-speed SAR ship detection. Remote
Sens. 2019,11, 2483. [CrossRef]
44.
Wang, Y.; Wang, C.; Zhang, H.; Dong, Y.; Wei, S. A SAR dataset of ship detection for deep learning under complex backgrounds.
Remote Sens. 2019,11, 765. [CrossRef]
45.
Chang, Y.L.; Anagaw, A.; Chang, L.; Wang, Y.C.; Hsiao, C.Y.; Lee, W.H. Ship detection based on YOLOv2 for SAR imagery. Remote
Sens. 2019,11, 786. [CrossRef]
46.
Jiao, J.; Zhang, Y.; Sun, H.; Yang, X.; Gao, X.; Hong, W.; Fu, K.; Sun, X. A densely connected end-to-end neural network for
multiscale and multiscene SAR ship detection. IEEE Access 2018,6, 20881–20892. [CrossRef]
47.
Li, J.; Qu, C.; Shao, J. Ship detection in SAR images based on an improved faster R-CNN. In Proceedings of the 2017 SAR in Big
Data Era: Models, Methods and Applications (BIGSARDATA), Beijing, China, 13–14 November 2017 ; pp. 1–6.
48.
Wei, S.; Zeng, X.; Qu, Q.; Wang, M.; Su, H.; Shi, J. HRSID: A high-resolution SAR images dataset for ship detection and instance
segmentation. IEEE Access 2020,8, 120234–120254. [CrossRef]
49.
Colin, A.; Fablet, R.; Tandeo, P.; Husson, R.; Peureux, C.; Longépé, N.; Mouche, A. Semantic segmentation of metoceanic processes
using SAR observations and deep learning. Remote Sens. 2022,14, 851. [CrossRef]
50.
Biffia, A.; Dodaro, L.; Morabito, F.; Pergolizzi, A. Mare Monstrum. 2020. Available online: https://www.legambiente.it/wp-
content/uploads/Dossier-Mare-Monstrum-2019.pdf (accessed on 13 December 2022).
51.
Biffia, A.; Dodaro, L.; Morabito, F.; Pergolizzi, A. Mare Monstrum. 2019. Available online: https://www.legambiente.it/wp-
content/uploads/2020/06/MareMonstrum2020.pdf (accessed on 13 December 2022).
52.
Furesi, R.; Madau, F.; Pulina, P.; Sai, R.; Pinna, M.; Pais, A. Profitability and sustainability of edible sea urchin fishery in Sardinia
(Italy). J. Coast. Conserv. 2016,20, 299–306. [CrossRef]
53.
Spera, G.; Monaco, P. Rapporto Annuale sul Controllo Della Pesca in Italia. 2019. Available online: https://www.guardiacostiera.
gov.it/stampa/Documents/RAPPORTO%20ANNUALE%202020%20-%20DOWNLOAD.pdf (accessed on 13 December 2022).
54.
Graziano, M.D.; Prete, R.D.; Grasso, M. A New Dataset of Multi-Mission/multi-Frequency sar Data for Maritime Monitoring:
First Results and Critical Analysis. 2022. Available online: https://iafastro.directory/iac/paper/id/74577/summary/ (accessed
on 13 December 2022).
55.
Miler, R.K. The Exactearth Satelite-AIS Benefits for the Maritime Transport. Prace Naukowe Wy˙
zszej Szkoły Bankowej w Gda´nsku
2013,28, 71–86.
56.
Dellepiane, S.; De Laurentiis, R.; Giordano, F. Coastline extraction from SAR images and a method for the evaluation of the
coastline precision. Pattern Recognit. Lett. 2004,25, 1461–1470. [CrossRef]
57.
Open Street Map. “Data Derived from OpenStreetMap for Download”, Open Street Map [Online]. 2022. Available online:
https://osmdata.openstreetmap.de/data/land-polygons.html (accessed on 7 December 2022).
58.
Benachenhou, K.; Taleb-Ahmed, A.; Hamadouche, M. Performances evaluation of GNSS ALTBOC acquisition with CFAR
detection in Rayleigh fading channel. In Proceedings of the 2013 Saudi International Electronics, Communications and Photonics
Conference, Riyadh, Saudi Arabia, 27–30 April 2013; pp. 1–7.
59.
Guida, M.; Longo, M.; Lops, M. Biparametric CFAR procedures for lognormal clutter. IEEE Trans. Aerosp. Electron. Syst.
1993
,
29, 798–809. [CrossRef]
60.
Anastassopoulos, V.; Lampropoulos, G.A. Optimal CFAR detection in Weibull clutter. IEEE Trans. Aerosp. Electron. Syst.
1995
,
31, 52–64. [CrossRef]
61.
Erfanian, S.; Vakili, V.T. Introducing excision switching-CFAR in K distributed sea clutter. Signal Process.
2009
,89, 1023–1031.
[CrossRef]
62.
Leng, X.; Ji, K.; Yang, K.; Zou, H. A Bilateral CFAR Algorithm for Ship Detection in SAR Images. IEEE Geosci. Remote Sens. Lett.
2015,12, 1536–1540. [CrossRef]
Disclaimer/Publisher’s Note:
The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.
... As part of the COAST project (COastal Area monitoring with SAR data and multimission/multi-frequency Techniques), funded by the Italian Space Agency (ASI), a dataset was constructed comprising stacks of SAR images from different missions, with minimal temporal lag below 15 minutes. Specifically, the dataset [10] encompasses multi-mission products with diverse acquisition modes and polarization. COSMO-SkyMed products are characterized by single HH polarization, while Sentinel-1 and SAOCOM products offer dual polarization (VH+VV) [11], [12], [13], [14]. ...
... It is worth noting that the inclusion of multi-frequency SAR data in the context of ship detection has received limited attention in the remote sensing literature. The dataset of [10] was not initially intended for DL algorithms. This paper aims to curate and annotate it to suit the requirements of an Object Detection (OD) architecture. ...
... The selection of these scenarios specifically targets activities related to illegal fishing, smuggling, and unauthorized trespass in Marine Protected Areas (MPAs). A complete description of the matching strategy adopted and selected scenarios of interest is reported in [10]. The number and types of products selected are reported in Table II. ...
... For a detected target R n , if there exists a true target G m in {G m } such that the IoU between R n and G m is greater than 0.5, R n is a correctly detected target; otherwise, it is a false alarm. The detection rate is defined as the ratio of the number of correctly detected true targets, M t , to the number of true targets, M, and the false rate is defined as the ratio of the number of false alarm detected targets, N f , to the number of detected targets, N [40], i.e., ...
Article
Full-text available
The detection of harbors presents difficulties related to their diverse sizes, varying morphology and scattering, and complex backgrounds. To avoid the extraction of unstable geometric features, in this paper, we propose an unsupervised harbor detection method for polarimetric SAR images using context features and polarimetric reflection symmetry. First, the image is segmented into three region types, i.e., water low-scattering regions, strong-scattering urban regions, and other regions, based on a multi-region Markov random field (MRF) segmentation method. Second, by leveraging the fact that harbors are surrounded by water on one side and a large number of buildings on the other, the coastal narrow-band area is extracted from the low-scattering regions, and the harbor regions of interest (ROIs) are determined by extracting the strong-scattering regions from the narrow-band area. Finally, by using the scattering reflection asymmetry of harbor buildings, harbors are identified based on the global threshold segmentation of the horizontal, vertical, and circular co- and cross-polarization correlation powers of the extracted ROIs. The effectiveness of the proposed method was validated with experiments on RADARSAT-2 quad-polarization images of Zhanjiang, Fuzhou, Lingshui, and Dalian, China; San Francisco, USA; and Singapore. The proposed method had high detection rates and low false detection rates in the complex coastal environment scenarios studied, far outperforming the traditional spatial harbor detection method considered for comparison.
... L-band and P-band SAR, which have longer wavelengths, have relatively longer penetration depth into vegetation and soil than shorter wavelengths (e.g., X-and C-band) (Ouchi, 2013). Combining multi-band SAR satellite data allows us to take advantage of multiple missions currently available, and the potential of multi-band SAR data has been demonstrated in a variety of applications (Del Prete et al., 2023;Fu et al., 2023). Previous DSM studies mainly focused on the potential of single-band SAR data, especially C-band Sentinel-1, in DSM applications (Tripathi and Tiwari, 2021;Zhang et al., 2023), and rarely investigated the effects of SAR data at different frequencies on soil prediction models. ...
... The Team has also developed innovative analytical models for relative motion design, applicable to both FF (e.g., to enable Distributed Synthetic Aperture Radar applications, DSAR) and missions scenarios involving close-proximity operations. Concerning OOS and ADR, which require relative navigation skills with respect to passively cooperative (i.e., equipped with fiducial markers designed to be easily detected and identified within the raw data produced by an electro-optical sensor) and non-cooperative space targets, the Team has developed innovative techniques based on data processing from passive (monocular and stereo cameras) and active (Lidar and Time of [82][83][84][85][86][87][88][89][90] . ...
Chapter
Full-text available
This chapter summarizes the research activities and main outcomes of the groups engaged in the fields of Aerospace Engineering, in the decade 2013–2023. The research topics are Aerodynamics, Aerospace Systems, Design of Aircraft and Flight, Fluid-dynamics, Propulsion and Structural Technologies.
... Expanding the training dataset is a potential approach to enhance the generalization ability and robustness of a deep-learningbased model. However, it should be noted that this method is not without its challenges, as data collection is a labor-intensive and time-consuming task that requires significant financial and material support to obtain a large number of labeled data [10]. Furthermore, it is important to recognize that increasing the dataset is not a panacea for improving the model's robustness and generalization ability. ...
Article
Full-text available
Ship detection in the maritime domain awareness field has seen a significant shift towards deep-learning-based techniques as the mainstream approach. However, most existing deep-learning-based ship detection models adopt a random sampling strategy for training data, neglecting the complexity differences among samples and the learning progress of the model, which hinders training efficiency, robustness, and generalization ability. To address this issue, we propose a ship detection model called the Leap-Forward-Learning-Decay and Curriculum Learning-based Network (LFLD-CLbased NET). This model incorporates innovative strategies as Leap-Forward-Learning-Decay and curriculum learning to enhance its ship detection capabilities. The LFLD-CLbased NET is composed of ResNet as the feature extraction unit, combined with a difficulty generator and a difficulty scheduler. The difficulty generator in LFLD-CLbased NET effectively expands data samples based on real ocean scenarios, and the difficulty scheduler constructs corresponding curriculum training data, enabling the model to be trained in an orderly manner from easy to difficult. The Leap-Forward-Learning-Decay strategy, which allows for flexible adjustment of the learning rate during curriculum training, is proposed for enhancing training efficiency. Our experimental findings demonstrate that our model achieved a detection accuracy of 86.635%, approximately 10% higher than other deep-learning-based ship detection models. In addition, we conducted extensive supplementary experiments to evaluate the effectiveness of the learning rate adjustment strategy and curriculum training in ship detection tasks. Furthermore, we conducted exploratory experiments on different modules to compare performance differences under varying parameter configurations.
... With increasing maritime activities and frequent vessel movements, the large-scale information on vessel activities poses an urgent challenge for the maritime surveillance system [4][5][6][7]. Current existing traditional vessel monitoring systems have security risks and cannot meet the demands of the maritime industry [8][9][10]. ...
Article
Full-text available
Vessel monitoring technology involves the application of remote sensing technologies to detect and identify vessels in various environments, which is critical for monitoring vessel traffic, identifying potential threats, and facilitating maritime safety and security to achieve real-time maritime awareness in military and civilian domains. However, most existing vessel monitoring models tend to focus on a single remote sensing information source, leading to limited detection functionality and underutilization of available information. In light of these limitations, this paper proposes a comprehensive ship monitoring system that integrates remote satellite devices and nearshore detection equipment. The system employs ResNet, a deep learning model, along with data augmentation and transfer learning techniques to enable bidirectional detection of satellite cloud images and nearshore outboard profile images, thereby alleviating prevailing issues such as low detection accuracy, homogeneous functionality, and poor image recognition applicability. Empirical findings based on two real-world vessel monitoring datasets demonstrate that the proposed system consistently performs best in both nearshore identification and remote detection. Additionally, extensive supplementary experiments were conducted to evaluate the effectiveness of different modules and discuss the constraints of current deep learning-based vessel monitoring models.
Article
Many operationally available satellite X-band SAR missions (TerraSAR-X, Kompsat-5, and COSMO-SkyMed) and new missions based on microsatellite constellations in X-band SAR (ICEYE and Capella) have dramatically increased maritime monitoring capabilities with shorter revisit times. Vessel detection in constellated SAR images has been based on various detection algorithms. AIS has been widely used to match the positions of vessels detected by SAR and identify dark vessels. However, the quality assessment of small-vessel detection has rarely been studied, and the requirements for additional monitoring data to supplement AIS have not been qualified in conjunction with vessels detected by SAR. In this study, the possibility of tracking vessels based on the constellation of SAR images was tested using an integrated field survey designed to capture small-vessel operations during SAR image acquisition from the constellations of Kompsat-5, TerraSAR-X, and Capella. The movement of ships with length of 10 m and 20 m could be estimated if the ships are commonly detected in SAR images within a short time difference. The necessity of the V-pass as additional monitoring data to supplement the AIS was quantitatively tested by comparing the matching rates of ships detected in SAR and AIS with and without the V-pass. When the V-pass data were added to the AIS, the number of vessels for the ground truth within the SAR image boundary increased and the false alarms were decreased. In the near future, the combined use of a vessel monitoring system and the integration of increasing constellations of X-band SAR images will help fill data gaps for marine monitoring, including illegal fishing vessels and marine oil spills.
Article
Full-text available
Through the Synthetic Aperture Radar (SAR) embarked on the satellites Sentinel-1A and Sentinel-1B of the Copernicus program, a large quantity of observations is routinely acquired over the oceans. A wide range of features from both oceanic (e.g., biological slicks, icebergs, etc.) and meteorologic origin (e.g., rain cells, wind streaks, etc.) are distinguishable on these acquisitions. This paper studies the semantic segmentation of ten metoceanic processes either in the context of a large quantity of image-level groundtruths (i.e., weakly-supervised framework) or of scarce pixel-level groundtruths (i.e., fully-supervised framework). Our main result is that a fully-supervised model outperforms any tested weakly-supervised algorithm. Adding more segmentation examples in the training set would further increase the precision of the predictions. Trained on 20 × 20 km imagettes acquired from the WV acquisition mode of the Sentinel-1 mission, the model is shown to generalize, under some assumptions, to wide-swath SAR data, which further extents its application domain to coastal areas.
Article
Full-text available
Azimuth ambiguities, or ghosts on SAR images, represent one of the main obstacles for SAR applications involving coastal monitoring activities such as ship detection. While most previous methods based on azimuth antenna pattern and direct filtering are effective for azimuth ambiguity suppression, they may not be effective for fast cruising small ships. This paper proposes a unique approach for the reduction of azimuth ambiguities or ghosts in SAR single-look complex (SLC) images using a simple rotation matrix. It exploits the fact that the signal powers of azimuth ambiguities are concentrated on narrow bands, while those of vessels or other true ground targets are dispersed over broad bands. Through sub-aperture processing and simple axis rotation, it is possible to concentrate the dispersed energy of vessels onto a single axis while the ghost signal powers are dispersed onto three different axes. Then, the azimuth ambiguities can be easily suppressed by a simple calculation of weighted sum and difference, while preserving vessels. Applied results achieved by processing TerrSAR-X SLC images are provided and discussed. An optimum weight of 0.5 was determined by Receiver Operating Characteristic (ROC) analysis. Capabilities of ship detection from the test image were significantly improved by removing 93% of false alarms. Application results demonstrate its high performance of ghost suppression. This method can be employed as a pre-processing tool of SAR images for ship detection in coastal waters.
Article
Full-text available
Spaceborne synthetic aperture radar (SAR) represents a powerful source of data for enhancing maritime domain awareness (MDA). Wakes generated by traveling vessels hold a crucial role in MDA since they can be exploited both for ship route and velocity estimation and as a marker of ship presence. Even if deep learning (DL) has led to an impressive performance boost on a variety of computer vision tasks, its usage for automatic target recognition (ATR) in SAR images to support MDA is still limited to the detection of ships rather than ship wakes. A dataset is presented in this paper and several state-of-the-art object detectors based on convolutional neural networks (CNNs) are tested with different backbones. The dataset, including more than 250 wake chips, is realized by visually inspecting Sentinel-1 images over highly trafficked maritime sites. Extensive experiments are shown to characterize CNNs for the wake detection task. For the first time, a deep-learning approach is implemented to specifically detect ship wakes without any a-priori knowledge or cuing about the location of the vessel that generated the wake. No annotated dataset was available to train deep-learning detectors on this task, which is instead presented in this paper. Moreover, the benchmarks achieved for different detectors point out promising features and weak points of the relevant approaches. Thus, the work also aims at stimulating more research in this promising, but still under-investigated, field.
Article
Full-text available
Recently, international agencies for border security ask for an improvement of the actual Maritime Situational Awareness. This manuscript presents preliminary results of a detection technique of go-fast boats, whose utilization in illegal affairs is strongly increasing. Their detection is very challenging since: (i) their echo is not visible in SAR images, and (ii) the illegal activities are carried out in the nighttime making useless the optical sensors. However, their wakes are very persistent and extent in SAR images for some kilometers. Hence, the manuscript shows an innovative deterministic methodology for the ship detection based on the wake signature. It firstly identifies pixels crossed by the wakes, whose presence is, then, validated in two steps. The first level of validation estimated how prominent the wake components are with respect to their background. The second level of validation exploits the presence of the wakes among neighbor pixels. The approach has been applied on ships imaged by TerraSAR-X mission showing the same peculiarities of go-fast boats. Results highlight the potentialities of the proposed approach, which can be also conceived as a subsequent step in a hybrid system, whose preliminary wake detection screening is carried out by different techniques.
Article
Full-text available
With the development of satellite technology, up to date imaging mode of synthetic aperture radar (SAR) satellite can provide higher resolution SAR imageries, which benefits ship detection and instance segmentation. Meanwhile, object detectors based on convolutional neural network (CNN) show high performance on SAR ship detection even without land-ocean segmentation; but with respective shortcomings, such as the relatively small size of SAR images for ship detection, limited SAR training samples, and inappropriate annotations, in existing SAR ship datasets, related research is hampered. To promote the development of CNN based ship detection and instance segmentation, we have constructed a High-Resolution SAR Images Dataset (HRSID). In addition to object detection, instance segmentation can also be implemented on HRSID. As for dataset construction, under the overlapped ratio of 25%, 136 panoramic SAR imageries with ranging resolution from 1m to 5m are cropped to 800 × 800 pixels SAR images. To reduce wrong annotation and missing annotation, optical remote sensing imageries are applied to reduce the interferes from harbor constructions. There are 5604 cropped SAR images and 16951 ships in HRSID, and we have divided HRSID into a training set (65% SAR images) and test set (35% SAR images) with the format of Microsoft Common Objects in Context (MS COCO). 8 state-of-the-art detectors are experimented on HRSID to build the baseline; MS COCO evaluation metrics are applicated for comprehensive evaluation. Experimental results reveal that ship detection and instance segmentation can be well implemented on HRSID.
Article
Full-text available
As an active microwave imaging sensor for the high-resolution earth observation, synthetic aperture radar (SAR) has been extensively applied in military, agriculture, geology, ecology, oceanography, etc., due to its prominent advantages of all-weather and all-time working capacity. Especially, in the marine field, SAR can provide numerous high-quality services for fishery management, traffic control, sea-ice monitoring, marine environmental protection, etc. Among them, ship detection in SAR images has attracted more and more attention on account of the urgent requirements of maritime rescue and military strategy formulation. Nowadays, most researches are focusing on improving the ship detection accuracy, while the detection speed is frequently neglected, regardless of traditional feature extraction methods or modern deep learning (DL) methods. However, the high-speed SAR ship detection is of great practical value, because it can provide real-time maritime disaster rescue and emergency military planning. Therefore, in order to address this problem, we proposed a novel high-speed SAR ship detection approach by mainly using depthwise separable convolution neural network (DS-CNN). In this approach, we integrated multi-scale detection mechanism, concatenation mechanism and anchor box mechanism to establish a brand-new light-weight network architecture for the high-speed SAR ship detection. We used DS-CNN, which consists of a depthwise convolution (D-Conv2D) and a pointwise convolution (P-Conv2D), to substitute for the conventional convolution neural network (C-CNN). In this way, the number of network parameters gets obviously decreased, and the ship detection speed gets dramatically improved. We experimented on an open SAR ship detection dataset (SSDD) to validate the correctness and feasibility of the proposed method. To verify the strong migration capacity of our method, we also carried out actual ship detection on a wide-region large-size Sentinel-1 SAR image. Ultimately, under the same hardware platform with NVIDIA RTX2080Ti GPU, the experimental results indicated that the ship detection speed of our proposed method is faster than other methods, meanwhile the detection accuracy is only lightly sacrificed compared with the state-of-art object detectors. Our method has great application value in real-time maritime disaster rescue and emergency military planning.
Article
Full-text available
The synergic utilization of data from different sources, either ground-based or spaceborne, can lead to effective monitoring of maritime activities. To this end, the integration of synthetic aperture radar (SAR) images with data reported by the automatic identification system (AIS) is of high interest. Accurate matching of ships detected in SAR images with AIS data requires compensation of the azimuth offset, which depends on the ship’s velocity. The existing procedures interpolate the route information gathered by AIS to estimate the ship’s velocity at the epoch of the SAR data, to remove the offset. Matching accuracy is limited by interpolation errors and AIS route information unavailability or uncertainties. This paper proposes the use of SAR-based ship velocity estimations to improve the integration of AIS and SAR data. A case study has been analyzed, in which the method has been tested on TerraSAR-X images collected over the Gulf of Naples, Italy. Presented results show that the matching is improved with respect to standard procedures. The proposed method limits the distance between the AIS report and the SAR-based detection to less than 150 m, which is in line with maritime surveillance needs.
Chapter
Why is it crucial to understand microwave synthetic aperture radar (SAR) theories? They have a tremendous role in comprehending the mechanisms of radar object imaging. Mineral and hydrocarbon explorations using synthetic aperture radar are mainly based on image processing filter techniques. Numerous academic institutions are using the SAR as a mapping tool without understanding the microwave theories behind its operation. The restriction of commercial software creates a such narrow SAR operational technique in the field of mineral and hydrocarbon explorations. The foremost aim of this chapter is to illuminate the significant theories behind microwave SAR. It well reveals that the radar equation is based on the perfect understanding of the Maxwell's equations. For instance, radar polarization cannot be addressed without prior knowledge in how electromagnetic waves propagate and interact act, for instance, with geological structural.
Conference Paper
Tracking of ships in an open sea scenario is a key maritime surveillance requirement. Ship tracking using airborne radar platforms is not very well addressed in the literature. In this paper, a concept of ship tracking using single-channel range-compressed (RC) airborne radar data is proposed. The proposed tracking algorithm is suitable for dense multi-target scenarios. Tracking is performed in the range-Doppler domain where the moving ships may appear out of the clutter region, improving their detectability. Ship tracking is an extended target tracking problem, therefore the center of gravity of the detected ship pixels are tracked over longer time. A powerful track management system is developed us-ing SQLite database which can handle gaps in the data and also the false detections. Some simulations and real experi-mental results from DLR’s F-SAR system are provided to prove the concept.
Article
This paper proposes a multi-scale rotation-invariant haar-like (MSRI-HL) feature integrated convolutional neural network (MSRIHL-CNN)-based ship detection algorithm of the multiple-target environment in synthetic aperture radar (SAR) imagery. Usually, ship detection includes preprocessing, prescreening, discrimination, and classification. Among them, prescreening and discrimination are the most two important stages so that they catch great intention. Based on our previous work, we propose a truncated-clutter-statistics-based joint, constant false alarm rate (CFAR) detector (TCS-JCFAR) for ship target prescreening in the multiple-target environment. TCS-JCFAR greatly enhances the prescreening rate in the multiple-target environment while achieving a low observed FAR. In the discrimination stage, conventional CNN extracts the deep features (high-level features); however, it will lose the local texture and edge information (low-level features) which are of great significance for target discrimination. Hence, the MSRI-HL features are used to represent the multi-scale, rotation-invariant texture, and edge information that conventional CNN fails to capture. The extracted low-level MSRI-HL features and the high-level deep features are optimally fused to a multi-layered feature vector. Finally, the multi-layered feature vector is fed into a typical support vector machine (SVM) classifier for ship target discrimination. The proposed MSRIHL-CNN combines the low-level texture and edge features and the high-level deep features; moreover, they are optimally fused to fully represent the ship targets. Undoubtedly, MSRIHL-CNN has better discrimination performance. The superiority of the proposed TCS-JCFAR-based prescreener and MSRIHL-CNN-based discriminator is validated on the Chinese Gaofen-3 SAR imagery.